Computational Pathology and Improving Predictive Analytics
32
Midwest Consortium for Computational Pathology Workshop January 26 th , 2021 Computational Pathology and Improving Predictive Analytics Michael J. Becich, MD PhD - [email protected] Chairman and Distinguished University Professor, Department of Biomedical Informatics http://www.dbmi.pitt.edu University of Pittsburgh School of Medicine
Transcript of Computational Pathology and Improving Predictive Analytics
Midwest Consortium for Computational Pathology Workshop
January 26th, 2021
Computational Pathology andImproving Predictive Analytics
Michael J. Becich, MD PhD - [email protected] and Distinguished University Professor,
Department of Biomedical Informaticshttp://www.dbmi.pitt.edu
University of Pittsburgh School of Medicine
Disclosures of COI for 2021 for MJBStartup/Public Companies Royalties/Licensing, Equity):
– De-ID Data Corp – de-identification software (licensing agreement) http://www.de-idata.com/
– Nexi – Newco by Rebecca Jacobson/TIES/TCRN team (royalties to my Department) to support pharma/biotech – http://www.nexihub.com
– SpIntellx (formerly SpDX) – Spatial Intelligence for Cancer Diagnostics (founder equity)
Consultancy (honoraria)– Cancer Center Consulting – Baylor, CINJ/Rutgers, U Colorado, U NM– CTSA Consulting – MCW, Northwestern, UC Davis, U Chicago, U IN, UC
Davis, U NM, U WI– Biomedical Informatics Consulting – Northwestern, Rockefeller, UC Davis, U
Chicago, U FLFederal Grants
– CDC NIOSH – National Mesothelioma Virtual Bank– NCI - CCSG – Cancer Bioinformatics Shared Facility– NCATS - CTSA and ACT– NLM - BMI TP– PCORI – PaTH CRN
Disclaimer: I am a member of NCI’s Board of Scientific Advisors and Frederic National Labs Advisory Board Tech WG
Take Home Messages• FDA approval of whole slide imaging is driving
Computational Pathology and AI in Pathology• Informatics is key to impact of Comp Path &
predictive analytics in clinical practice• Pathology and Radiology partnerships need to be
enabled• Key Impact Areas – Predictive Analytics for health
and discovery science via Comp Path!
Whole Slide Imaging (WSI) Workflow
From Bertram and Kopfleish, Vet Path 2017
FDA Approval for Whole Slide Imaging
From Pantanowitz, Path Info Summit, 2017
Computational Pathology – What is it?Computational Path = Big Data Science Meets Pathology• Massive Increases in Volume of Digital Data Generated
from WSI PLUS genomic sequencing data –• Heralds the rise of computational pathology• Critical for Personalized Medicine, Learning Health
Systems, Basic Research and “Big Data/Data Science”
From Fuchs, 2017
Introduction to Whole Slide Imaging
From Saltz, NLM circa 2011
Definition of Computational Pathology• An approach to diagnosis that
incorporates multiple sources of data (H&E, IHC, IF & genomic data)
• Presents clinically actionable knowledge (big data to knowledge)
• Advanced decision support for precision (personalized) medicine
• Helps to redefine Pathology from an observational to knowledge engineering discipline hence critical to healthcare data science (Louis et al Arch Path Lab Med 2014)
Data Types Critical to Success for Predictive Analytics• Imaging – Rich source of computable information
– Need to De-ID WSI and then “deeply” annotate• Phenotype – From Anatomic Pathology Lab Info Sys
– Structured data from synoptic reports– Unstructured data via NLP (Text Info Extract Sys – TIES)
• Computational Pathology Annotation – more later• Outcomes Data – From Cancer Registry Systems• Data to Integrate – Biobanks, Clinical Pathology &
Molecular Pathology
Support for Comp Path – Integration Needed
What is the Text Info Extraction System (TIES)?• An NLP and Information Retreival system for
de-identifying, annotating, storing and retrieving pathology and radiology documents
• A system for indexing research resources (FFPE, FF, WSI) with document annotations
• An GUI for querying large repository of annotated documents and obtaining resources locally, using an honest broker model
• A platform to support data, biospecimen and both Pathology and Radiology images for sharing among networks of cancer centers and other institutions
TCRN Pubs
http://ties.dbmi.pitt.edu/
Causal Network Discovery = Computational PathologyA probabilistic network approach to uncover genetic drivers of melanoma using data on copy number variation and gene expression*
Akavia UD , et al. Cell 143 (2010) 1005-1017.(The figure above appears in this paper)
Presenter
Presentation Notes
In this research, the investigators developed and applied a causal-network-discovery algorithm that uses genomic data (in the form of copy number variation) along with gene expression data to identify those genes that are likely to be causing (driving) melanoma in human patients. Abstract from the cited paper: Systematic characterization of cancer genomes has revealed a staggering number of diverse aberrations that differ among individuals, such that the functional importance and physiological impact of most tumor genetic alterations remain poorly defined. We developed a computational framework that integrates chromosomal copy number and gene expression data for detecting aberrations that promote cancer progression. We demonstrate the utility of this framework using a melanoma data set. Our analysis correctly identified known drivers of melanoma and predicted multiple tumor dependencies. Two dependencies, TBC1D16 and RAB27A, confirmed empirically, suggest that abnormal regulation of protein trafficking contributes to proliferation in melanoma. Together, these results demonstrate the ability of integrative Bayesian approaches to identify candidate drivers with biological, and possibly therapeutic, importance in cancer.
Fact
s pe
r Dec
isio
n
1000
10
100
5Human Cognitive
Capacity
Computational Pathology Algorithms like those in CCD for imaging and genomics are key enablers!!!
2000 20101990 2020
Structural Genetics: e.g. SNPs, haplotypes
Functional Genetics: Gene expression
profiles
Proteomics and othereffector molecules
Decisions by Clinical Phenotype
From William Stead: http://courses.mbl.edu/mi/2009/presentations_fall/SteadV1.ppt & http://www.mbl.edu/education/courses/special_topics/pdf/med_sched09_fall.pdf
1. Intra-tumoral spatial heterogeneity
17Nature Reviews Cancer (2009) 9:239–252
1. Quantify Heterogeneity for Diagnostics, Prognostics & Immunotherapy
18Marusyk A. Nat Rev Cancer. 2012 Apr 19;12(5):323-34
2. Integration of H&E, IHC and IF
19
H&E in transmitted light Multi to hyperplexed
• Universal method • Limited molecular measurements
in transmitted (IHC)• Complex tissue “scenes”
• Emerging method with potential to measure many DNA/RNA/proteins in the same tissue section/TMA
• Structural biomarkers• Quantify biomarker expression
levels and tissue based spatial relationships.
3. Histopatholomics: Spatial ITH
EpigenomicsGenomics
ProteomicsMetabolomics
1 mm
Multi to hyperplexed fluorescence imaging of whole section for higher spatial resolution and tissue context
H&E stained whole tissue sectionfrom FFPE tumor sample
1
4
4 2A
23
13
Thanks to Dr. Joe Ayoob for HPlomics
4. Ground-truth & Annotation
21
5. ”Callable” ID of Histologic Features
6. Triaging ROIs
23
7. Computational Pathology and AI
24
Error over time
Intervention
Pathology and Radiology Must Partner!
From Fuchs, 2017
End of Talk – e-mail me at [email protected] if you have questions/clarifications not covered in the discussion.
NOTE: E-mail me if you want PDFs of articles or presentation.
Thank you for organizing this MCCP Workshop!
Slide 27
Association for Pathology Informatics (API)http://www.pathologyinformatics.org
“…to advance the field of pathology informatics as an academic and a clinical subspecialty of pathology…”
Journal of Pathology InformaticsCo-Editors Liron Pantanowitz, MD PhD and Anil Parwani, MD PhD
http://www.jpathinformatics.org
Please support JPI, API and
Pathology Informatics as the Home for Digital
Pathology -Great Academic
and Strategic Partnership with
Multiple Benefits!!!
Thanx to the Pitt team!
http://www.csb.pitt.edu/comppath/
30
Youtube Channel for Lectures
https://www.youtube.com/watch?v=Xrvxc0YNcAM&feature=youtu.be
Thanx to Chakra Chennubhotla, PhD & Burak Tosun, PhD
Comp Path Team Profile - Pittsburgh• Special Interest Group Members (n=93):
– Bioengineering (5%)– Biomedical Informatics (10%)– Computational and Systems Biology (10%)– Computer Science and Machine Learning (20%)– Industry and Entrepreneurs (20%)– Medicine (10%)– Pathology Informatics (10%)– UPMC (15%)Participation from Carnegie Mellon & Dusquene Universities
Computational Pathology Lecture Archive -https://www.youtube.com/channel/UCWfBS3PLWHTIACeccm2CIog
![[0.4cm] Scaling Limits in Computational Bayesian Inversion · PDF fileScaling Limits in Computational Bayesian Inversion Claudia Schillings, Christoph Schwab Warwick Centre for Predictive](https://static.fdocuments.in/doc/165x107/5a883e367f8b9aa5408e6d85/04cm-scaling-limits-in-computational-bayesian-inversion-limits-in-computational.jpg)
